The combination of artificial neural networks and synchrotron radiation-based infrared micro-spectroscopy for a study on the protein composition of human glial tumors.

Protein-related changes associated with the development of human brain gliomas are of increasing interest in modern neuro-oncology. It is due to the fact that they might make some of these tumors highly aggressive and difficult to treat. This paper presents a methodology for protein-based analysis of human brain gliomas using synchrotron radiation based Fourier transform infrared spectroscopy (SRFTIR) coupled with artificial neural networks (ANNs). The main goal of this study was to optimize a set of ANNs to predict the secondary structure of proteins (alpha-helices, beta-sheets, beta-turns, bends, random coils) in brain gliomas, based on the amide I-II spectral range. All networks were tested and optimized to reach the standard error of prediction (SEP) lower than 5%. The results indicate that protein-related changes are associated with a tumor's malignancy grade. Particularly, the content of alpha helices increases with increasing malignancy grade, while the content of beta sheets decreases. We also found that proteomic information could be a useful marker to distinguish either between low and high grade tumors or between oligodendroglial- and astrocyte-derived ones. This demonstrates the applicability of FTIR coupled with ANNs to provide clinically relevant information.